work.

A collection of work I have done for the past while that I’m proud of.

A backlog of unfinished ideas can be found here.

---

writing.

You can find internal monologue under posts index.

open source.

• Quartz - 🌱 a fast, batteries-included static-site generator that transforms Markdown content into fully functional websites (2023-)

  • A set of tools that helps you publish your digital garden and notes as a website for free.
  • Improved performance of graph interaction with Canvas jackyzha0/quartz#1328
  • Added support for PDF in popover modal jackyzha0/quartz#913
  • Implemented font-fetching before runtime jackyzha0/quartz#817
  • Implemented telescope-style search jackyzha0/quartz#722, jackyzha0/quartz#774, jackyzha0/quartz#782
  • Added sidenotes components, inspired by Tuffe’s CSS jackyzha0/quartz#1555, examples
  • Landing page of this website, with custom components, i.e: supper club, curius, parsing jupyter notebooks
  • Source: https://github.com/jackyzha0/quartz and site

• avante.nvim - A Cursor-like chat IDE for Neovim (2024-)

  • Implemented bounding UI popover to improve QOL yetone/avante.nvim#29
  • Added support for lazy setup for better load time improvement yetone/avante.nvim#14
  • Added Rust crates for .avanterules templates https://yetone/avante.nvim/pull/466
  • Source: https://github.com/yetone/avante.nvim

• tinymorph - An exploration into how we build interfaces for machine-assisted writing tool (2024-)

  • WARNING: Currently in research phase.
  • Trained sparse autoencoder to interpret Llama 3.2 features (Templeton et al., 2024)

• OpenLLM - Run any open-source LLMs as OpenAI compatible API endpoint in the cloud. (2023-)

  • 🔬 Build for fast and production usages
  • 🚂 Support Llama, Qwen, Gemma, etc, and quantized versions
  • ⛓️ OpenAI-compatible API
  • 💬 Built-in ChatGPT like UI
  • 🔥 Accelerated LLM decoding with state-of-the-art inference backends
  • Source: https://github.com/bentoml/openllm

• BentoML - Build Production-grade AI Application (2021-) (Yang et al., n.d.)

  • a framework that simplifies machine learning model deployment and provides a faster way to ship your model to production. Supports a variety of use cases, from classical ML to LLMs, diffusions models.
  • Built using Python, BuildKit, gRPC
  • Source: https://github.com/bentoml/bentoml, Documentation

• incogni.to - a pseudonymous event platform that curates for those yearning to be seen for who they are, not what they can "sell" (2024)

  • Implemented a RAG pipeline for recommendation system based on users preferences and interests, with command-r-plus-08-2024, deployed with vLLM and BentoML (Yang et al., n.d.)
  • Added semantic search to find relevant events based on query with Cohere Rerank
  • General UI implementation with shadcn/ui and vercel/next.js
  • Demoed at New Build’24
  • Source: stream, posts

• onw - A real-time navigation tools for safer commute (2021)

  • Implemented route optimization, heat map visualization to identify hot zones, peer notification system.
  • Added a heuristic Gaussian Mixture Model to find the safest path between different locations, trained on past assault data provided by Toronto Police Department.
  • Awarded: Finalists at Hack the North 2021.
  • Built using AWS Fargate, React Native, TypeScript, GraphQL, Apache Spark MLlib, Google Maps API
  • Source: https://github.com/tiproad/omw, devpost

talks.

• OpenLLM, and everything about running LLMs in production at Hack The North (2023)
  • Source: slides

companies.

Im thinking to build a toronto compute company, looking for funding

— aaron (@aarnphm_) 11 octobre 2024

References

• Templeton, A., Conerly, T., Marcus, J., Lindsey, J., Bricken, T., Chen, B., Pearce, A., Citro, C., Ameisen, E., Jones, A., Cunningham, H., Turner, N. L., McDougall, C., MacDiarmid, M., Freeman, C. D., Sumers, T. R., Rees, E., Batson, J., Jermyn, A., … Henighan, T. (2024). Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet. Transformer Circuits Thread. [link]
• Yang, C., Sean, S., Aaron, P., Shenyang, Z., Sauyon, L., Bo, J., Fog, D., Xipeng, G., & Frost, M. (n.d.). BentoML: The framework for building reliable, scalable and cost-efficient AI application. [GitHub]